ORIGINAL PAPER

Differential and combined impacts of extreme temperaturesand air pollution on human mortality in south–centralCanada. Part I: historical analysis

Chad Shouquan Cheng & Monica Campbell & Qian Li & Guilong Li & Heather Auld &

Nancy Day & David Pengelly & Sarah Gingrich & Joan Klaassen & Don MacIver &

Neil Comer & Yang Mao & Wendy Thompson & Hong Lin

Received: 30 June 2008 /Accepted: 23 January 2009 /Published online: 18 February 2009# The Author(s) 2009. This article is published with open access at Springerlink.com

Abstract This paper forms the first part of an introductionto a synoptic weather typing approach to assess differentialand combined impacts of extreme temperatures and airpollution on human mortality in south–central Canada,focusing on historical analysis (a companion paper—Part IIfocusing on future estimates). In this study, an automatedsynoptic weather typing procedure was used to identifyweather types that have a marked association with high airpollution levels and temperature extremes, and facilitatesassessments of the differential and combined health impactsof extreme temperatures and air pollution. Annual meanelevated mortality (when daily mortality exceeds thebaseline) associated with extreme temperatures and acuteexposures to air pollution, based on 1954–2000, was 1,082

[95% confidence interval (CI) of 1,017–1,147] for Mon-treal, 1,047 (CI 994–1,100) for Toronto, 462 (CI 438–486)for Ottawa, and 327 (CI 311–343) for Windsor. Of thisannual mean elevated mortality, extreme temperatures areusually associated with roughly 20%, while air pollution isassociated with the remaining 80%. Three pollutants(ozone, sulfur dioxide, and nitrogen dioxide) are associatedwith approximately 75% of total air pollution-relatedmortality across the study area. The remaining 25% isalmost evenly associated with suspended particles andcarbon monoxide, the other two pollutants addressed inthis study. Of the five pollutants, ozone is most significantlyassociated with elevated mortality, making up one-third ofthe total air pollution-related mortality. PM2.5 and PM10

were not used as a measure of particulate in the study dueto brief data records. The study results also suggest that, onthe basis of daily mortality risks, extreme temperature-related weather presents a much greater risk to humanhealth during heat waves and cold spells than air pollution.For example, in Montreal and Toronto, daily mean elevatedmortality counts within the hottest weather type were twiceas high as those within air pollution-related weather types.

Keywords Temperature . Air pollution . Humanmortality .

Synoptic weather typing . South–central Canada

Introduction

It is well known that air pollution has been strongly linkedto human health problems, particularly in vulnerablepopulations such as the elderly, young children, and thosesuffering from cardio-respiratory conditions (World HealthOrganization 2004). For example, Judek et al. (2004)

Air Qual Atmos Health (2008) 1:209–222DOI 10.1007/s11869-009-0027-1

C. S. Cheng (*) :Q. Li :G. Li : J. Klaassen :N. Comer :H. LinAtmospheric Science and Applications Unit,Meteorological Service of Canada Branch—Ontario,Environment Canada,4905 Dufferin Street,Toronto, ON, Canada M3H 5T4e-mail: shouquan.cheng@ec.gc.ca

M. Campbell :N. Day : S. GingrichToronto Public Health,Toronto, Canada

H. Auld :D. MacIverAdaptation and Impacts Research Division,Science and Technology Branch, Environment Canada,Toronto, Canada

D. PengellyDepartment of Medicine, McMaster University,Hamilton, Canada

Y. Mao :W. ThompsonPublic Health Agency of Canada,Ottawa, Canada

pointed out that the annual excess number of deathsassociated with short-term and long-term exposure to airpollution was estimated to be 1,800 and 4,200, respectively,for eight Census Divisions in Canada with a population of8.9 million (national total population of 30.0 million in the2001 Census). Another report released by Toronto PublicHealth (2004) estimated that about 1,700 premature deathseach year in the City of Toronto (population=2.5 million inthe 2001 Census) were associated with acute or short-termexposures to ozone (O3), nitrogen dioxide (NO2), carbonmonoxide (CO), sulfur dioxide (SO2), and chronic or long-term exposures to fine particulate matter (PM2.5). About700 of these deaths were attributable to acute exposuresalone. The Ontario Medical Association (OMA 2005)estimated that, based on acute exposure to O3, SO2, NO2,CO, and PM, air pollution causes approximately 5,800premature deaths per year in Ontario (population=11.4 mil-lion in the 2001 Census) and almost $8 billion in directcosts to the provincial health care system.

It is also well known that temperature extremes areresponsible for significant numbers of deaths during hotand cold days. For example, a recent study (Medina-Ramónand Schwartz 2007) investigated the relationships betweentemperature extremes and mortality in 50 US cities,resulting in mortality increases associated to both extremeheat (+5.74%) and extreme cold (+1.59%). In the US,during the hot summer of 1980, an estimated 10,000 deathswere related to oppressive heat (Ross and Lott 2003). Thesevere 5-day heat wave over the central United Statesduring mid-July 1995 caused 830 deaths, with 525 of thesedeaths in the City of Chicago (Changnon et al. 1996).During August 2003, western European countries experi-enced a record-breaking, heat-wave event; in France aloneover 14,800 deaths may have been directly caused by thisheat wave (Pirard et al. 2005).

Many chronic diseases, especially those of the respira-tory and cardiovascular systems, are exacerbated mostfrequently during or after a period of specific weatherconditions (e.g., Kalkstein 1991; Cheng 1991; Nichols et al.1995; Sheridan and Kalkstein 2004; Shin et al. 2008) and/or high air pollution concentrations (Schwartz et al. 1991;Burnett et al. 1994, 1999; Schwartz 2000, Basu and Ostro2008). As shown in Table 1 for examples, a number of theprevious studies in this field have examined how human-made factors such as atmospheric pollutants affect humanhealth; many others have focused on how natural stressorssuch as extreme temperatures influence death rates. Someother studies (e.g., Rainham and Smoyer-Tomic 2003; Huet al. 2008) have investigated associations between extremetemperatures and mortality, considering confoundingimpacts of air pollutants. However, the differential andcombined impacts of extreme temperature-related weatherevents and high air pollution episodes on human health are

poorly understood. In the light of these concerns, Environ-ment Canada, in partnerships with Toronto Public Health,McMaster University, Health Canada, and Public HealthAgency of Canada, has completed a 3-year research project.This project, funded by the Health Policy ResearchProgram, Health Canada, proposed a method based onsynoptic weather typing in addition to air pollution levels toevaluate differential and combined impacts of extremetemperatures and air pollution on human mortality in south–central Canada. The results from this study, expressed aselevated mortality (above the baseline) associated withdifferent factors (e.g., heat, cold, pollutants), could be usableby governmental agencies and stakeholders for assisting indeveloping better policies on health protection and balancingpolicy decisions. This study may enhance current understand-ing of environmental problems related to human health insouth–central Canada.

The current paper describes the background to thedevelopment of an analysis approach to quantitativelyestimate elevated mortality, applying an automated synopticweather typing and mortality baseline analyses. Thesynoptic weather typing has been adapted to develop aheat–health watch/warning system piloted in Philadelphia,Cincinnati, Rome, Shanghai, and Toronto for the UNShowcase Project (Sheridan and Kalkstein 2004). Thismethod promises to evaluate relationships among a varietyof weather elements rather than only individual variables—the synergistic impact of several elements being moresignificant than the sum of their individual impacts(Kalkstein 1991; Pope and Kalkstein 1996; Kalkstein etal. 1997). Part of the current study focusing on climatechange impacts analysis on both future air pollution- andtemperature extreme-related mortality in south–centralCanada is presented in a companion paper (Part II, Chenget al. 2009). To effectively estimate climate change-inducedfuture elevated mortality, historical analysis is essential forus to build a solid science foundation on health impacts ofair pollution and extreme temperatures. For this study, fourcities, Montreal, Ottawa, Toronto, and Windsor, located insouth–central Canada, were selected (Fig. 1). There areseveral reasons for selection of the cities. First, the cities,especially for Windsor, have experienced the hottestweather situation in the country. Second, Toronto andMontreal are the two largest cities, in terms of population,in Canada. Finally, Ottawa is the capital of Canada.

Data sources and treatment

To evaluate differential and combined impacts of airpollution and extreme temperatures on human mortality,meteorological, air pollution, and mortality data are needed.The detailed information, regarding data sources, variables,

210 Air Qual Atmos Health (2008) 1:209–222

Tab

le1

Examples

ofprevious

stud

ieson

health

impactsof

airpo

llutio

n/extrem

etemperature

Study

Metho

dology

Major

find

ings

Study

area

Airpo

llutio

nandhu

man

health

Schwartz

etal.19

91Poisson

regression

Acute

respiratoryillness

increasesby

27–28%

whenTSP

orNO2levelsincrease

from

10to

70μg/m

3Fivecities,Germany

Dockery

etal.19

92Poisson

regression

Mortalityincrease

by16

–17%

per10

0μg/m

3increase

inPM

10

TwoUScities

Stiebet

al.19

96A

regression

similarto

Poisson

regression

Asthm

aem

ergencyvisits:33%

higher

whendaily

maxim

umO3>75

ppb

Saint

John

,Canada

Burnettet

al.19

97Rando

meffectsrelativ

erisk

regression

O3andrespiratoryho

spitalizations,with

relativ

erisks:1.00

0–1.08

816

Canadiancities

Burnettet

al.19

99Generalized

additiv

emod

els

Per

10μg/m

3increase

inPM

2.5/PM

10,therelatedcardio-respiratory

hospitalizations:3.3%

/1.9%

more

Toron

to,Canada

Goldb

erget

al.20

03Generalized

linearPoisson

regression

andnaturalsplines

Daily

mortalityincreasesby

4.32

%/4.08%

dueto

increase

insuspended

particles/NO2,across

inter-quartilerange

Mon

treal,Canada

Fun

get

al.20

05Generalized

linearPoisson

regression

&naturalsplin

esDaily

hospitalizations

increase

by2.6%

/4.0%/5.6%

for1-/2-/3-daymeanSO2

increase

across

inter-qu

artilerang

eWindsor,Canada

Keatin

ge&

Don

aldson

2006

Generalized

additiv

emod

els

Pollutantsplayed

little

partin

excess

mortalityassociated

with

hotweather

Lon

don,

UK

Qianet

al.20

08Generalized

additiv

emod

els

Daily

mortalityincreasesby

2.20

%,3.28

%,2.35

%forno

n-accidental,

cardiovascular,stroke

per10

μg/m

3increase

inPM

10du

ring

hottemperature

Wuh

an,China

Shinet

al.20

08Generalized

additiv

ePoisson

mod

els

Pub

lichealth

risksrelatedto

NO2appear

tobe

increasing

buttherisks

forO3isweak

12Canadiancities

Won

get

al.20

08Poisson

regression

andnaturalsplin

esExcessmortalityrisk

per10

μg/m

3relatedto

NO2:0.90

–1.97%

,SO2:

0.87–1

.61%

,PM

10:0.26–1

.25%

,O3:0.31–0

.63%

Fou

rcitiesin

Asia

Extremetemperature

andhu

man

health

Kalkstein

1991

Syn

optic

weather

typing

Heatbu

tno

tpo

llutantsplayed

mostpartin

excess

mortality

St.Lou

is,USA

Kalkstein

&Greene19

97Syn

optic

weather

tying

Avery

warm

summer

airmassof

maritimeorigin

canincrease

daily

mortalityas

manyas

30deaths

inlargeeasternUScities

44UScities

Guestet

al.19

99Thresho

ldtemperature:Tmax>28

°CAnn

ualmeanheat-related

mortalityis17

5Fivecities,Australia

Dessai20

02Non

-linearregression

Heat-relatedmortalityissixper10

0,00

0individu

als

Lisbo

n,Portugal

Daviset

al.20

03Thresho

ldtemperature:16

:00LSTheat

index>30

°CHeat-relatedmortalityrate

permillionpeop

lewas

41.0

in19

60s–19

70s,

17.3

in19

80s,10

.5in

1990

s28

cities,USA

McG

rego

ret

al.20

04Linearregression

Ann

ualischaemic

heartdiseasemortalitypeak

occurs

during

thecoldest

winterseason

sFivecoun

ties,UK

Basu&

Ostro

2008

Tim

e-stratifiedcase-crossov

erapproach

Cardiov

ascularmortalityincreasesby

2.6%

per4.7°Cincrease

inmean

daily

apparent

temperature

California,

USA

Huet

al.20

08Tim

e-series

classificatio

nandregression

tree

(CART)

Daily

mortalityincreasesby

7.3%

whenTmax>32

°Cand12

.1%

when

meanSO2exceeds0.31

5pp

mSyd

ney,

Australia

Rocklöv

&Forsberg20

08Generalized

additiv

ePoisson

mod

els

Daily

mortalityincreasesby

3.1–

7.7%

during

heat

waves

Stockho

lm,Sweden

Air Qual Atmos Health (2008) 1:209–222 211

and record length, is described in Table 2. In this study,hourly surface meteorological data for the period 1954–2000 were used for the treatment of missing data. Missingdata of each meteorological variable were interpolatedusing a temporal linear method in cases where the datawere missing for three consecutive hours or less; days withdata missing for four or more consecutive hours wereexcluded from the analysis. Of the total dataset, about 1%of the total days required missing data interpolation; afterinterpolation, the dataset was over 99.7% complete.Additionally, only six-hourly weather variables observed

at 03:00, 09:00, 15:00, and 21:00 local standard time (LST)were used in the study. For air pollution data, severalmonitoring sites listed in Table 2, based on the length andcompleteness of the available data records, were chosen tocalculate hourly/daily mean concentrations representingaverage air pollution conditions for each of the selectedcities (refer to Cheng et al. 2007a for details on air pollutiondata treatment).

To effectively evaluate extreme temperature- and airpollution-related health risks, the daily aggregated non-traumatic mortality data have to be treated to remove inter-

Table 2 Data used in the study

Description Variable Source City

Total non-traumatic mortality (1954–2000):

ICD-6 (1953–1957): 001–795 Daily aggregated mortality counts Statistics Canada Montreal (Island)

ICD-7 (1958–1968): 001–795 Ottawa

ICD-8 (1969–1978): 000–796 Toronto (Metropolitan)

ICD-9 (1979–1999): 001–799 Windsor

ICD-10 (2000): A00–R99

Six-hourly surface meteorologicalobservations at 03:00, 09:00,15:00, 21:00 local standardtime (1954–2000)

Air temperature (°C) Environment Canada’s NationalClimate Data andInformation Archive

International airports inMontreal, Ottawa, Toronto,and Windsor

Dew point (°C)

Sea-level air pressure (hPa)

Total cloud cover (tenths)

U-wind and V-wind (m s−1)

Hourly/daily air pollution observations:

1980–2000 for ozone Carbon monoxide (CO) Environment Canada’s NationalAir Pollution Surveillance(NAPS) network

Monitoring stations:

1974–1996 for suspended particles Nitrogen dioxide (NO2) Montreal: 7

1974–2000 for rest of pollutants Ozone (O3) Ottawa: 1

Sulfur dioxide (SO2) Toronto: 3

Suspended particles (SP) Windsor: 2

Fig. 1 Location map of the fourselected cities

212 Air Qual Atmos Health (2008) 1:209–222

annual trends and seasonal fluctuations. The inter-annualtrends over time for total non-traumatic mortality are shownin Fig. 2. These upward trends likely resulted from severalfactors, including increases in total population and changesin population age structure (Kalkstein et al. 1997). Otherfactors (i.e., improvements to health care and economic/living conditions) could reduce the upward trends. Sincethis study attempts to analyze health impacts of extremetemperatures and air pollution, such non-environmentalfactors, representing confounding influences within themortality data, should be removed. The mortality upwardtrends over time can be removed by fitting a polynomialregression line through annual mean mortality countsagainst time (in years), which is the baseline I shown inFig. 2. It is noteworthy that the upward trend pattern forMontreal is different from that of other cities; this is a resultof a decline in population (about 0.2 million) from 1971 to1981, based on the census data of Statistics Canada. Theresidual or difference between each day’s actual mortalitycount and the baseline count (Fig. 2) indicates that upward

trends due to non-environmental factors have been re-moved. From removing non-environmental factors usingthe baseline I, some health impacts of extreme temperaturesand air pollution were also removed since the extreme hotand polluted days were included in construction of thebaseline I. The next step is to modify the baseline I todevelop the baseline II using only weather types associatedwith relatively good air quality and comfortable weatherconditions (see the section ‘Results and discussion’ for adetailed information).

In addition to inter-annual trends, the mortality seasonalfluctuations need to be removed from the analysis. Usingmonthly mortality baseline II, seasonal fluctuations can beexcluded from mortality data by removing monthly inter-annual trends. The monthly mortality baseline II wasdeveloped by fitting monthly polynomial regression mod-els, which usually explain over 80% of the data variancesfor the four selected cities. In addition to inter-annual andseasonal fluctuations of mortality, consideration was givento removing weekly fluctuations of mortality from the data.

Windsor: y = 1.94E-05x3 - 0.0026x2 + 0.1208x + 3.011R2 = 0.9521 RMSE = 0.1243

Ottawa: y = 0.1627x + 4.48 R2 = 0.9848 RMSE = 0.28

0

2

4

6

8

10

12

14

1950 1955 1960 1965 1970 1975 1980 1985 1990 1995 2000Year

An

nu

al M

ean

Mo

rtal

ity

Ottawa Windsor

Montreal: y = 1.97E-04x3 - 0.0171x2 + 0.5828x + 30.877R2 = 0.8372 RMSE = 1.095

Toronto: y = 5.79E-05x3 - 0.0079x2 + 0.6614x + 26.116R2 = 0.9797 RMSE = 0.8247

20

25

30

35

40

45

50

1950 1955 1960 1965 1970 1975 1980 1985 1990 1995 2000Year

An

nu

al M

ean

Mo

rtal

ity

Montreal Toronto

Fig. 2 Inter-annual total non-traumatic mortality trends(baseline I). The x described inequations is set as 1 for the year1954, 2 for 1955, and so on

Air Qual Atmos Health (2008) 1:209–222 213

However, the weekly fluctuations within a weather groupwere not statistically significant for the four selected citiesand therefore were not considered to be removed.

Analysis techniques

To analyze differential and combined impacts of extremetemperatures and air pollution on human health, three majormethods or steps are used in the study. First, an automatedsynoptic weather typing was employed to classify each dayof the dataset into a distinctive weather type based onweather characteristics represented by a variety of weathervariables. Second, daily air pollution data were used toidentify the weather types most highly associated with highair pollution levels for each of the pollutants. Finally,temperature extreme- and air pollution-related mortalitycould be distinguished by linking daily elevated mortality(above the baseline) with hot/cold/air pollution-relatedsynoptic weather types. The detailed information onanalysis techniques is summarized in Fig. 3.

Synoptic weather typing

An automated synoptic weather typing, based primarily onair mass similarity and differentiation within and betweenweather types, was used to assign each day of the dataset toa distinctive weather type. The entire suite of 24 weathervariables, including six-hourly surface weather observa-tions of air temperature, dew point temperature, sea-levelair pressure, total cloud cover, and south–north and west–east winds, were used in synoptic weather typing. As partof this research, Cheng et al. (2007a, 2007b) havedeveloped weather typing procedures for evaluation ofclimate change impacts on air pollution concentrations forthe selected four cities, which were also used in this study.Due to space limitation, weather typing procedures arebriefly summarized here. Weather typing procedures are

comprised of a two-step process: (1) daily synoptic weathertypes were categorized using principal components analysis(PCA) and a hierarchical agglomerative cluster method—anaverage linkage clustering procedure, and (2) all days in thedataset were regrouped by a non-hierarchical method—discriminant function analysis using the centroids of thehierarchical weather types as seeds. First, a hierarchicalagglomerative clustering procedure was suitable for theinitial classification since the number of weather types wasnot predetermined (Davis 1986; Kalkstein et al. 1996;Cheng and Kalkstein 1997; Cheng and Lam 2000). Thecorrelation matrix-based PCA (Jolliffe 1986; Applequist etal. 2002) was performed to reduce the 24 inter-correlatedweather variables into seven linearly independent compo-nents, explaining much of the variance (i.e., 88–92% acrossthe cities) within the original dataset. Days with similarmeteorological conditions tended to exhibit similar compo-nent scores. The average linkage clustering proceduregenerated statistical diagnostics to produce an appropriatenumber of clusters. The diagnostics attempts to minimizethe within-cluster variances and to maximize the between-cluster variances (Boyce 1996; Cheng and Lam 2000;Cheng et al. 2007a). Second, following the hierarchicalclassification, a non-hierarchical classification procedurewas used to reclassify all days within the dataset.Improvement in the cluster structure resulting from non-hierarchical reclassification was evaluated by comparingwithin-/between-cluster variances. Cheng et al. (2007a)have concluded that the reclassification results were betterthan the original, with smaller within-cluster variances andlarger between-cluster variances, which is consistent withthat of previous studies (e.g., DeGaetano 1996).

Link between weather types and air quality

Following weather type classification, daily mean andmaximum air pollution concentrations were used to identifythe synoptic weather types most highly associated with high

Six-hourly Surface Weather Data(1954–2000)

Hourly/Daily Air Pollution Data

(1974–2000)

Missing Data TreatmentData Treatment:

Calculating Daily Mean Air Pollution Concentration Within a City

Synoptic Weather Typing:Principal Components AnalysisClustering ProcedureDiscriminant Function Analysis

Identification of Hot/Cold/Air Pollution-related Weather Types

Evaluation of Hot-/Cold-/Air Pollution-related Health Risks

Daily Mortality Data (1954–2000)

Data Treatment:Mortality Baseline to Remove

Inter-annual Trends and Seasonal Fluctuation

Fig. 3 Flow chart of methodol-ogies and steps used in the study

214 Air Qual Atmos Health (2008) 1:209–222

air pollution concentrations for each of the selectedpollutants. As shown in Table 3, two measurements,within-weather-type mean air pollution concentrations andfrequency of high air pollution episodes, were used todetermine relationships between weather types and airquality. Daily mean air pollution concentrations for eachtype were then determined to ascertain whether the airpollution levels within the particular weather types weredistinctively high or low. In addition, a ratio of the within-weather-type occurrence of high air pollution episodes(actual frequency) to the occurrence of the weather type inthe entire record (expected frequency) was utilized to

determine whether any of the weather types were over-represented for high air pollution episodes. Weather typeswith ratios significantly greater than 1.0 clearly had agreater proportion of days with high air pollution episodesthan would be expected. If the frequency ratio was greaterthan 1.0 and the within-weather-type mean concentrationwas higher than the overall mean, that weather type wasselected as an air pollution-related weather type. As shownin Table 3 for Toronto, nine weather types were defined asthe O3-related types. A χ2 test was used to ascertainwhether the actual frequencies of high air pollution events(daily air pollution concentration with one, two, and three

Table 3 Within-weather-type O3 level by concentration anomalies (overall mean=17.0 ppb) for nine O3-related weather types (types 1–9) andexamples of nine low-O3-level types (types 17–25) in Toronto (1980–2000)

Synoptic type O3 concentration level

Mean2

(ppb)1 std. dev. above mean 2 std. dev. above mean 3 std. dev. above mean

Type Days Freq.(%)1 (a)

Days Freq.(%)3 (b)

Ratio4

(b)/(a)Days Freq.

(%)3 (c)Ratio4

(c)/(a)Days Freq.

(%)3 (d)Ratio4

(d)/(a)

1 173 2.26 32.02 130 11.76 5.21 74 18.78 8.31 21 19.09 8.45

2 223 2.91 26.14 136 12.31 4.23 62 15.74 5.40 23 20.91 7.18

3 260 3.40 28.47 186 16.83 4.96 83 21.07 6.20 23 20.91 6.16

4 233 3.04 19.77 115 10.41 3.42 45 11.42 3.75 12 10.91 3.59

5 161 2.10 21.87 88 7.96 3.79 37 9.39 4.47 12 10.91 5.19

6 107 1.40 15.94 40 3.62 2.59 12 3.05 2.18 5 4.55 3.25

7 81 1.06 15.30 35 3.17 2.99 8 2.03 1.92 1 0.91 0.86

8 348 4.54 5.19 64 5.79 1.27 12 3.05 0.67 1 0.91 0.20

9 285 3.72 8.36 64 5.79 1.56 13 3.30 0.89 1 0.91 0.24

17 270 3.53 −13.68 0 0.00 0.00 0 0.00 0.00 0 0.00 0.00

18 235 3.07 −11.72 0 0.00 0.00 0 0.00 0.00 0 0.00 0.00

19 125 1.63 −11.54 1 0.09 0.06 0 0.00 0.00 0 0.00 0.00

20 138 1.80 −15.99 1 0.09 0.05 0 0.00 0.00 0 0.00 0.00

21 70 0.91 −11.29 0 0.00 0.00 0 0.00 0.00 0 0.00 0.00

22 304 3.97 −15.70 1 0.09 0.02 0 0.00 0.00 0 0.00 0.00

23 110 1.44 −9.95 0 0.00 0.00 0 0.00 0.00 0 0.00 0.00

24 203 2.65 −16.80 2 0.18 0.07 0 0.00 0.00 0 0.00 0.00

25 174 2.27 −11.04 0 0.00 0.00 0 0.00 0.00 0 0.00 0.00

Sub-total5 2,114 27.61 18.31 973 88.05 3.19 385 97.72 3.53 110 100 3.62

Total6 7,655 1,105 394 110

1 Percentage occurrence of the weather type or expected frequency for high O3-level events (within-weather-type number of days divided by thenumber of total days)2Within-weather-type daily mean anomalies of O3 concentrations (overall mean=17.0 ppb)3 Percentage occurrence (actual frequency) of the days with one, two, and three standard deviations above the overall mean (17.0 ppb) within aparticular weather type (within-weather-type number of the events divided by the number of total events)4 Ratio of actual frequency of the high O3-level events over the expected frequency. A ratio greater than one indicates that a larger proportion ofdays in the weather type possess higher O3 levels than would be expected5 The sum of the nine O3-related weather types and some small weather types that are most associated with high O3 concentration levels, but notshown in Table 46 The sum of all weather types

Air Qual Atmos Health (2008) 1:209–222 215

standard deviations above the overall mean) were signifi-cantly different from their expected occurrences (χ2 testsignificance level of 0.001). This method was then appliedto different pollutants for each of the selected cities.

Health impacts of extreme temperatures and air pollution

Synoptic weather typing can be used to evaluate thedifferential and combined impacts of extreme temperaturesand air pollution on elevated mortality. In this study, theelevated mortality counts within the weather types associ-ated with high air pollution concentrations—but withcomfortable weather conditions—are treated as the impactsof air pollution on human health. The impacts of extremetemperatures on human health might be determined by thehot and cold weather types. Although the hot/cold weathergroups are usually associated with high air pollutionconcentrations, elevated mortality within the hot/coldgroups was defined as extreme temperature-related mortal-ity since extreme temperatures are most relevant to elevatedmortality in these weather groups (refer to the section‘Results and discussion’ for a more detailed information).

The synoptic weather typing procedure can be used toanalyze differential impacts of air pollutants on elevatedmortality counts as well. The air pollution-related weathertypes may be associated with high air pollution concen-trations for one single pollutant or multiple pollutants. Forthe former case, within-weather-type elevated mortality canbe identified as associated with that particular pollutant. Forthe latter case, the study assumes that within-weather-typeelevated mortality is associated with the one particularpollutant that is most important to the weather type (refer tothe section ‘Results and discussion’ for a more detailedinformation).

Consideration was given to taking into account lag timesof extreme temperature and air pollution impacts onmortality. Based on the analysis, no-lag-time relationshipsbetween heat and mortality are more significant than otherlag times across the study area. This is consistent with theprevious studies (e.g., Dessai 2002; Rainham and Smoyer-Tomic 2003; Sheridan and Kalkstein 2004). As a result, thewithin-weather-type elevated mortality was estimated in thestudy based on lag-zero-day relationships.

Results and discussion

Findings derived from this study are significant in threeareas: (1) extreme temperature- and air pollution-relatedweather types, (2) mortality baseline, and (3) differentialand combined health impacts of temperature extremes andair pollution. The major findings in each of the areas arediscussed in this section as follows.

Extreme temperature- and air pollution-relatedweather types

The number of major synoptic weather types (with sizesabove 1% of the total days) identified for the selected citiesvaried slightly from one city to another: 39, 37, 36, and 39for Montreal, Ottawa, Toronto, and Windsor, respectively.The smaller weather types were still included in theanalysis. As part of this study, Cheng et al. (2007a) havedetermined extreme temperature- and air pollution-relatedweather types for the selected cities. For each of the fourselected cities, there are three hot weather types, four orfive cold weather types, and the number of air pollution-related weather types, as shown in Table 4 for Toronto(similar results were discovered for the other cities as well,but not shown). The three hot weather types (Hot1, Hot2,and Hot3) can capture 50–60%, 20–30%, and about 10% ofthe total days with 15:00 LST temperature ≥32°C,respectively, across the study area. For determination ofthe cold weather types, the criterion varies from location tolocation, based on the difference of January mean 15:00temperature among the cities. The within-weather-typemean 15:00 temperature of ≤−6°C was used to define acold weather type for Ottawa and Montreal, and of ≤−3°Cfor Toronto and Windsor.

As described in the previous section, air pollution-relatedweather types were defined based on (1) within-weather-typedaily mean concentrations and (2) the ratios of actualfrequency of high pollution level events over the expectedfrequency. The within-weather-type mean concentrationsand ratios for nine O3-related weather types and nine low-O3-level types in Toronto are shown in Table 3, as anexample. The similar results were found for other selectedcities but not shown due to limitations of space. Asshown in Table 3, daily mean anomalies of O3 concentra-tion within weather types 1–9 are much higher than theoverall mean (17.0 ppb in Toronto). Most of other weathertypes (e.g., weather types 17–25) possess negative meananomalies of O3 concentrations. For weather type 1 inToronto, the expected frequency of high air pollution eventswould be 2.26%, based on the size of the weather type.However, the actual frequency of the high air pollutionevents with one and three standard deviations above theoverall mean was about 12% and 19%, respectively—morethan five and eight times what might otherwise be expected.Through these analyses, ten, nine, nine, and nine O3-relatedweather types were identified for Montreal, Ottawa,Toronto, and Windsor, respectively, of which the first threeare also called hot weather types. On average across thefour selected cities, these weather types accounted for 83%and 97% of the total high O3 events with greater than oneand three standard deviations above the overall mean,respectively.

216 Air Qual Atmos Health (2008) 1:209–222

Similar analyses were also applied to the remainder ofthe pollutants for each of the selected cities. The summa-rized results of the weather types highly associated witheach of the pollutants are shown in Table 4 (using Torontoas an example). Some weather types showed significantlinkage with high air pollution concentrations for manypollutants (up to five); other types possessed good airquality and were not significantly related to any pollutant.If a weather type was associated with high air pollutionlevels for more than one pollutant, it was important todetermine which air pollutant was most significantly

associated with the weather type, based on the highestweather-type-frequency ratio described in the section(‘Analysis techniques’) above. For example, 16 weathertypes are associated with relative high SO2 levels inToronto; however, only five of them are most significantlyrelated to SO2, while for the rest of the weather types, otherpollutants are more important. As shown in Table 4,weather types for this study were divided into ten groups:three hot weather-related (including air pollution) groups(Hot1/AP, Hot2/AP, Hot3/AP), one cold weather-related(including air pollution) group (Cold/AP), five air

Table 4 Ten weather groups in Toronto: three hot, one cold, five air pollutant-related, and one “other”

Type 15:00 LST temperature(std. dev.)

15:00 LST dew point(std. dev.)

15:00 LST windspeed/direction

Pollutants associated with weathertypes

Weather groups

1 29.82 (2.73) 17.37 (3.21) 5.63 WSW O3 NO2 SP Hot1/AP

2 27.23 (3.11) 16.44 (3.42) 6.31 WSW O3 SO2 NO2 SP Hot2/AP

3 26.78 (2.63) 14.35 (3.38) 5.05 S O3 SO2 NO2 SP Hot3/AP

4 24.58 (3.04) 18.49 (2.16) 3.95 S O3 NO2 SP AP-O3

5 24.45 (3.04) 15.14 (3.39) 4.58 S O3 SO2 NO2 SP CO AP-O3

6 23.04 (2.91) 9.45 (3.74) 4.53 SSE O3 NO2 SP CO AP-O3

7 25.03 (2.35) 13.26 (2.97) 3.30 SE O3 AP-O3

8 19.31 (3.27) 11.99 (3.22) 3.78 ESE O3 NO2 AP-O3

9 21.03 (3.14) 7.64 (3.65) 4.08 SSE O3 NO2 CO AP-O3

10 16.64 (3.60) 11.95 (3.47) 4.25 SE NO2 AP-NO2

11 13.17 (3.74) 3.74 (3.52) 4.13 SE NO2 CO AP-NO2

12 0.92 (3.03) −3.46 (2.87) 5.90 W SO2 AP-SO2

13 8.73 (4.61) 0.63 (3.61) 5.27 W SO2 NO2 SP CO AP-SP

14 3.63 (3.55) −0.10 (3.14) 6.32 WSW SO2 SP CO AP-SO2

15 2.55 (4.75) −1.74 (4.14) 8.51 WSW SO2 AP-SO2

16 5.32 (3.90) 0.46 (3.59) 3.27 SE SO2 NO2 SP CO AP-NO2

17 7.71 (3.76) 4.23 (3.43) 3.78 SSE SO2 NO2 SP CO AP-SP

18 −2.48 (2.96) −5.67 (3.12) 5.22 ENE SO2 AP-SO2

19 16.23 (4.10) 10.53 (3.65) 5.54 SW SO2 NO2 SP CO AP-SP

20 12.75 (5.65) 2.96 (5.25) 4.27 SSE CO AP-CO

21 −1.10 (4.38) −8.29 (3.42) 5.27 NNW CO AP-CO

22 −2.87 (3.23) −8.03 (3.30) 3.81 SSW SO2 NO2 SP CO AP-SO2

23 −5.25 (3.39) −9.64 (3.67) 6.76 NNW Cold

24 −10.44 (3.55) −14.66 (3.83) 4.81 N SO2 NO2 SP CO Cold/AP

25 −9.77 (4.39) −16.03 (4.20) 4.06 NNW SO2 NO2 CO Cold/AP

26 −7.07 (3.55) −12.15 (3.71) 7.93 W SO2 Cold/AP

27 −2.97 (3.98) −7.60 (3.95) 5.16 SW SO2 SP CO Cold/AP

28 20.66 (3.33) 8.76 (3.50) 6.72 WNW Other

29 23.01 (3.68) 15.38 (3.00) 6.04 WNW Other

30 21.13 (3.09) 8.76 (3.14) 5.68 NNW Other

31 11.75 (4.03) −1.55 (3.67) 5.34 NW Other

32 7.79 (4.13) −0.51 (3.87) 7.29 WNW Other

33 3.37 (3.64) −1.97 (3.15) 5.28 NNW Other

34 2.71 (3.80) 0.42 (3.68) 6.55 E Other

35 2.39 (4.17) 0.44 (4.22) 5.23 N Other

36 −2.22 (4.38) −6.64 (4.39) 9.53 WNW Other

Air Qual Atmos Health (2008) 1:209–222 217

pollutant-related groups (CO, NO2, O3, SO2, SP), and one“other” group with relatively good air quality and comfort-able weather conditions. For each of the four cities beingstudied, the percentage occurrences of the ten weathergroups are fairly consistent across the study area, with onlya small variation (Fig. 4). Such weather grouping canfacilitate the analysis of differential and combined impactsof extreme temperatures and air pollution on humanmortality.

Mortality baseline

The inter-annual upward trends over time for total non-traumatic mortality (mortality baseline I) shown in Fig. 2were removed to minimize confounding influences of non-environmental factors (e.g., increases in total populationand changes in population age structure). However, toquantitatively assess elevated mortality (above the baseline)associated with extreme temperatures and air quality, themortality baseline I needs to be modified. Followingdetermination of the ten weather groups, mortality baselineI can be adjusted to develop baseline II using only the “other”weather group. In fact, some low-frequency weather types notshown in Table 4, which were identified as “other,” were alsoincluded in the development of baseline II. Baseline II wasdeveloped by calculating the within-“other”-weather-groupmean mortality from anomalous data against baseline I foreach 3-month season (December–February, March–May,June–August, September–November). As shown in Fig. 5for Toronto as an example (similar results were discovered inother cities), baseline II was lower than baseline I sincebaseline I includes all days (including hot/cold/high airpollution episodes) in its construction. The differencebetween two baselines that represents part of the extremetemperatures and air pollution effects, which was removedfrom baseline I, should be added back to the elevated

mortality. As a result, the baseline II can represent normal ornatural deaths without the effects of extreme temperaturesand air pollution. A positive mortality residual (abovebaseline II) should represent the elevated mortality associatedwith extreme temperatures and air pollution. A negativemortality residual (below baseline II) indicates a situationwith no elevated mortality relative to extreme temperaturesand air pollution. When within-weather-type annual anddaily mean elevated mortalities were calculated, the dailynegative mortality residual was set as zero.

The advantage of using the mortality baseline, as derivedfor the current study, is its enhancement of the relationshipbetween extreme temperatures/air pollution and mortality. Itcould also minimize the confounding impacts of non-environmental factors such as population increase andimprovements to health care and living/economic condi-tions. In the health studies, mortality rates were often usedto remove the impacts of population growth. When long-term health data (e.g., ten-year spans of data records ormore) are considered, it is insufficient to remove onlypopulation effects from mortality data so as to analyzehealth impacts of air pollution and extreme temperatures.To demonstrate this, population census data were used andtested to calculate mortality rates for the selected four cities.To effectively remove impacts of population increase andchanges in population age structure, an elderly population(≥65 years) was used. According to the results, thereappears to be a significant downward trend in mortalityrates for each of the cities, due to improvements to healthcare and living/economic conditions (e.g., use of airconditioning). On average, across the selected four cities,daily mean mortality rates per 100,000 elderly people wereabout 30 in the early 1950s, dropping to 15 by the year2000. For reliable, long-term, time-series health data, thisdownward trend should also be removed in order to preventthese non-environmental factors from influencing results.

12.0 11.2 9.7 9.1

16.7 16.9 17.6 18.1

5.7 8.6 6.6 6.06.1

7.6 7.4 8.2

12.7 8.0 9.9 8.2

10.8 9.113.5 15.6

28.1 30.926.6 24.0

2.1 2.4 2.3 2.34.12.82.73.2

4.43.62.62.7

0%

10%

20%

30%

40%

50%

60%

70%

80%

90%

100%

Montreal Ottawa Toronto Windsor

OT

SO2

NO2

SP

CO

O3

CA

HA3

HA2

HA1

Fig. 4 Percentage occurrence ofthe ten weather groups in thefour cities (HA1, HA2, and HA3represent the three hot weathergroups; CA is the cold weathergroup; O3, NO2, SO2, CO, andSP are different pollutant-relatedweather groups; and OT is the“other” or comfortable weathergroup)

218 Air Qual Atmos Health (2008) 1:209–222

Health impacts of temperature extremes and air pollution

Figure 6 shows quantitatively summarized results of thedifferential and combined impacts of extreme temperaturesand air pollution on elevated non-traumatic mortality for thefour selected cities. A mean annual total elevated mortalitywas calculated from the sum of positive mortality residualsfor all days within each of the ten weather groups, dividedby the total number of years. Generally, the proportion ofelevated mortality associated with extreme temperaturesand air pollution was consistent across the study area.Extreme temperatures were usually associated with over20% of mean annual total elevated mortality; air pollutionwas related to the remaining 80%. Of air pollution-related

mortality, three pollutants (O3, SO2, and NO2) wereassociated with about 75% across the study area. Theremaining 25% was almost evenly associated with SP andCO, the other two pollutants used in the study. Of the fivepollutants, O3 was the most highly associated with elevatedmortality in each of the cities, responsible for one-third ofthe total air pollution-related mortality.

Although “other” weather group is usually associatedwith relatively good air quality and comfortable weatherconditions, there still exists some degree of within-typevariance. The elevated mortality within the “other” weathergroup was still found to be associated with air pollution. Todemonstrate this, days within the “other” weather groupwere divided into air pollution concentration deciles foreach pollutant; the daily mean mortality was then calculatedfor each of the deciles. The results of the decile evaluationindicated that in the “other” weather group, there weresignificant relationships between air pollution concentra-tions and elevated mortality counts across the study area.Within-decile mean mortality residuals in the bottomdeciles were usually lower than those of the top decilesfor all pollutants except for CO. In addition to this decileanalysis, within-“other”-weather-group regression resultsalso show that only air pollutants other than weathervariables are significantly associated with elevated mortal-ity, with a model R2 of 0.71, 0.62, 0.40, and 0.30 forMontreal, Ottawa, Windsor and Toronto, respectively (referto a companion paper—Part II, Cheng et al. 2009 for detailson regression). This finding is consistent with that ofprevious studies (Lambert et al. 1998; Toronto PublicHealth 2001; Vedal et al. 2003), which concluded that

-8

-6

-4

-2

0

2

4

6

1954 1959 1964 1969 1974 1979 1984 1989 1994 1999

Year

Dai

ly M

ean

Sta

nd

ard

ized

Mo

rtal

ity Baseline I

Baseline II

Fig. 5 Sample of mortality baselines I and II for summer season(June–August) in Toronto

418

49

14

31

14

142

53 32 76 58

141 113

46 52

96 129

38

67 63

3920

71 60

35

25267 272

13476

38

4142

105

18218027

0

100

200

300

400

500

600

Tor Mon Ott Win Tor Mon Ott Win Tor Mon Ott Win

Num

ber o

f Ele

vate

d M

orta

lity

HA1 HA2 HA3 CA O3 SO2 NO2 CO SP OT

Heat and Cold Related Air Pollution Related Other (AP-related)

225263

94 69

555 546

234182

Toronto Annual Total = 1047Montreal Annual Total = 1082

Ottawa Annual Total = 462Windsor Annual Total = 327

22%24%

20%21%

53% 50%

51%

56%

29%

23%

25% 25%

131314

Fig. 6 Mean annual total elevated non-traumatic mortality caused byheat/cold and air pollutants in four cities (1954–2000). HA1, HA2, andHA3 represent three hot weather groups; CA is the cold weather group;O3, NO2, SO2, CO, and SP are different air pollutant-related weather

groups; and OT is the “other” weather group. Tor, Mon, Ott, and Winrepresent the four cities: Toronto, Montreal, Ottawa, and Windsor.Legend “left-to-right” is equivalent to bars “bottom-to-top”

Air Qual Atmos Health (2008) 1:209–222 219

there is no evidence of clear thresholds of pollutantconcentrations associated with increased elevated mortality.

In addition to annual total elevated mortality counts, thedaily mean elevated mortality breakdown by the tenweather groups was examined for the four cities. AsFig. 7 illustrates, daily mean elevated mortality was muchhigher for the extreme temperature-related weather groupsthan it was for either the air pollution-related or “other”weather groups, especially in Montreal and Toronto. Inthese two larger cities, daily mean elevated mortality countswithin the extreme hot weather group were twice as high asthose within “other” weather group. The smaller differencebetween daily elevated mortalities associated with extremetemperatures and air pollution in Ottawa and Windsormight be due to the small number of daily mortality counts.To more effectively analyze health impacts of extremetemperatures and air pollution for the smaller cities, moreresearch should be conducted using a larger daily samplesize of health outcomes, such as hospital admissions andemergency room visits.

Conclusions

A major goal of this study has been to analyze thedifferential and combined impacts of extreme temperatures

and air pollution on human mortality for four selected cities(Montreal, Ottawa, Toronto, and Windsor) in south–centralCanada. The major findings from this study were summa-rized as follows:

& The annual average number of elevated mortalityattributable to extreme temperatures and acute exposureto air pollution, based on 1954–2000, was 1,082 [95%confidence interval (CI) of 1,017–1,147] for Montreal,1,047 (CI 994–1,100) for Toronto, 462 (CI 438–486)for Ottawa, and 327 (CI 311–343) for Windsor.

& Of this elevated mortality, extreme temperatures wereusually associated with over 20% across the study area;air pollution was related to the remaining 80%.

& Of air pollution-related elevated mortality, three pollu-tants (O3, SO2, and NO2) were associated with about75% across the study area. The remaining 25% wasalmost evenly associated with SP and CO, the other twopollutants included in the study. Of the five pollutants,O3 was most highly associated with elevated mortalityin each of the cities, accounting for one-third of total airpollution-related mortality.

& Daily mean heat-related mortality was much higher thanthat associated with air pollution-related and otherweather groups, especially in Montreal and Toronto.Daily mean elevated mortality within the extreme hot

4.56

3.93

3.142.96

2.79 2.912.71

2.852.70 2.70

5.34

4.17

3.17 3.25

2.93 2.952.80

3.09

2.75 2.65

2

3

4

5

6

HA1 HA2 HA3 CA O3 SO2 NO2 CO SP OT

Dai

ly M

ean

Mo

rtal

ity

Toronto Montreal

0.95 0.93 0.91 0.960.87 0.91 0.91 0.92

0.85 0.87

1.441.37

1.54

1.301.23

1.361.30 1.26 1.29

1.19

0.0

0.2

0.4

0.6

0.8

1.0

1.2

1.4

1.6

1.8

HA1 HA2 HA3 CA O3 SO2 NO2 CO SP OT

Dai

ly M

ean

Mo

rtal

ity

Windsor Ottawa

Fig. 7 Daily mean elevatedmortality breakdown by weath-er groups: the top panel forToronto and Montreal and thebottom panel for Windsor andOttawa (1954–2000). HA1,HA2, and HA3 represent threehot weather groups; CA is thecold weather group; O3, NO2,SO2, CO, and SP are differentair pollutant-related weathergroups; and OT is the “other”weather group

220 Air Qual Atmos Health (2008) 1:209–222

weather group in Montreal (5.3; CI 4.7–5.9) andToronto (4.6; CI 4.1–5.1) was twice as high as thatassociated with “other” weather group (2.7; CI 2.5–2.9).

A general conclusion from this study is that a combina-tion of synoptic weather typing and mortality baselineapproaches can be useful to distinguish health impacts oftemperature extremes and air pollution concentrations. Thisstudy aims to provide decision makers with scientificinformation needed for public policy risk identificationand assessment. The results of this study, expressed aselevated mortality above the baseline associated withvarious factors (e.g., heat, cold, pollutants), can be usedby governmental agencies and stakeholders to help themdevelop better policies on health protection and to balancepolicy decisions. The study could enhance our understand-ing of environmental problems related to human health. Tosupport the recommendation of adaptation policies toreduce risks to vulnerable populations due to climatechange, it is necessary to estimate possible changes infuture extreme temperature- and air pollution-related mor-tality, which is the major objective of a companion paper(Part II, Cheng et al. 2009).

Acknowledgments This study was funded through the HealthPolicy Research Program, Health Canada (6795-15-2001/4400011).The views expressed herein are solely those of the authors and do notnecessarily represent the views or official policy of Health Canada orEnvironment Canada. The authors are grateful for the suggestionsmade by the Project Advisory Committee, which greatly improved thestudy. We also would like to thank two anonymous reviewers forproviding detailed comments that significantly improved the originalmanuscript.

Open Access This article is distributed under the terms of theCreative Commons Attribution Noncommercial License which per-mits any noncommercial use, distribution, and reproduction in anymedium, provided the original author(s) and source are credited.

References

Applequist S, Gahrs GE, Pfeffer RL, Niu XF (2002) Comparison ofmethodologies for probabilistic quantitative precipitation fore-casting. Weather Forecast 17:783–799. doi:10.1175/1520-0434(2002)017<0783:COMFPQ>2.0.CO;2

Basu R, Ostro BD (2008) A multicounty analysis identifying thepopulations vulnerable to mortality associated with high ambienttemperature in California. Am J Epidemiol 168:632–637.doi:10.1093/aje/kwn170

Boyce AJ (1996) Mapping diversity: a comparative study of somenumerical methods. In: Cole AJ (ed) Numerical taxonomy.Academic, New York, pp 1–30

Burnett RT, Bartlett S, Krewski D, Roberts G, Raad-Young M (1994)Air pollution effects on hospital admissions: a statistical analysisof parallel time series. Environ Ecol Stat 1:325–332.doi:10.1007/BF00469429

Burnett RT, Brook JR, Yung WT, Dales RE, Krewski D (1997)Association between ozone and hospitalization for respiratory

diseases in 16 Canadian cities. Environ Res 72:24–31.doi:10.1006/enrs.1996.3685

Burnett RT, Smith-Doiron M, Stieb D, Cakmak S, Brook J (1999)Effects of particulate and gaseous air pollution on cardiorespira-tory hospitalizations. Arch Environ Health 54:130–139

Changnon SA, Kunkel KE, Reinke BC (1996) Impacts and responsesto the 1995 heat wave: a call to action. Bull Am Meteorol Soc77:1497–1506. doi:10.1175/1520-0477(1996)077<1497:IARTTH>2.0.CO;2

Cheng CS (1991) Synoptic climatological categorization and humanmortality in Shanghai, China. Proc Middle S Div Assoc AmGeogr 24:5–11

Cheng CS, Campbell M, Li Q, Li G, Auld H, Day N, Pengelly D,Gingrich S, Yap D (2007a) A synoptic climatological approach toassess climate impact on air quality in south–central Canada. PartI: historical analysis. Water Air Soil Pollut 182:131–148.doi:10.1007/s11270-006-9327-3

Cheng CS, Campbell M, Li Q, Li G, Auld H, Day N, Pengelly D,Gingrich S, Yap D (2007b) A synoptic climatological approachto assess climate impact on air quality in south–central Canada.Part II: future estimates. Water Air Soil Pollut 182:117–130.doi:10.1007/s11270-006-9326-4

Cheng CS, Campbell M, Li Q, Li G, Auld H, Day N, Pengelly D,Gingrich S, Klaassen J, MacIver D, Comer N, Mao Y, ThompsonW, Lin H (2009) Differential and combined impacts of extremetemperatures and air pollution on human mortality in south–central Canada. Part II: future estimates. Air Qual Atmos Healthdoi:10.1007/s11869-009-0026-2

Cheng CS, Kalkstein LS (1997) Determination of climatologicalseasons for the East Coast of the U.S. using an air mass-basedclassification. Clim Res 8:107–116. doi:10.3354/cr008107

Cheng CS, Lam KC (2000) Synoptic typing and its application to theassessment of climatic impact on concentrations of sulfur dioxideand nitrogen oxides in Hong Kong. Atmos Environ 34:585–594.doi:10.1016/S1352-2310(99)00194-6

Davis JC (1986) Statistics and data analysis in geology, 2nd edn.Wiley, New York, p 646

Davis RE, Knappenberger PC, Michaels PJ, Novicoff WM (2003)Changing heat-related mortality in the United States. EnvironHealth Perspect 111:1712–1718

DeGaetano A (1996) Delineation of mesoscale climate zones in thenortheastern United States using a novel approach to clusteranalysis. J Clim 9:1765–1782. doi:10.1175/1520-0442(1996)009<1765:DOMCZI>2.0.CO;2

Dessai S (2002) Heat stress and mortality in Lisbon Part I. Modelconstruction and validation. Int J Biomet 47:6–12. doi:10.1007/s00484-002-0143-1

Dockery DW, Schwartz J, Spengler JD (1992) Air pollution and dailymortality: association with particulates and acid aerosols. EnvironRes 59:362–373. doi:10.1016/S0013-9351(05)80042-8

Fung KY, Luginaah I, Gorey KM, Webster G (2005) Air pollution anddaily hospital admissions for cardiovascular diseases in Windsor,Ontario. Can J Public Health 96:29–33

Goldberg MS, Burnett RT, Valois MF, Flegel K, Bailar JC III, BrookJ, Vincent R, Radon K (2003) Associations between ambient airpollution and daily mortality among persons with congestiveheart failure. Environ Res 91:8–20. doi:10.1016/S0013-9351(02)00022-1

Guest CS, Willson K, Woodward AJ, Hennessy K, Kalkstein LS,Skinner C, McMichael AJ (1999) Climate and mortality inAustralia: retrospective study, 1979–1990, and predicted impactsin five major cities in 2030. Clim Res 13:1–15. doi:10.3354/cr013001

Hu W, Mengersen K, McMichael A, Tong S (2008) Temperature, airpollution and total mortality during summers in Sydney, 1994–2004.Int J Biometeorol 52:689–696. doi:10.1007/s00484-008-0161-8

Air Qual Atmos Health (2008) 1:209–222 221

Jolliffe IT (1986) Principal component analysis. Springer, New York,p 271

Judek S, Jessiman B, Stieb D, Vet R (2004) Estimated number ofexcess deaths in Canada due to air pollution. http://www.hc-sc.gc.ca/ahc-asc/media/nr-cp/2005/2005_32bk2_e.html (accessedFebruary 2008)

Kalkstein LS (1991) A new approach to evaluate the impact of climateon human mortality. Environ Health Perspect 96:145–150.doi:10.2307/3431223

Kalkstein LS, Greene JS (1997) An evaluation of climate/mortalityrelationships in large U.S. cities and the possible impacts of aclimate change. Environ Health Perspect 105:84–93.doi:10.2307/3433067

Kalkstein LS, Nichols MC, Barthel CD, Greene JS (1996) A newspatial synoptic classification: application to air-mass analysis.Int J Climatol 16:983–1004. doi:10.1002/(SICI)1097-0088(199609)16:9<983::AID-JOC61>3.0.CO;2-N

Kalkstein LS, Barthel CD, Ye H, Smoyer K, Cheng CS, Greene JS,Nichols MC, Kalkstein AJ (1997) The impacts of weather andpollution on human mortality. Publ In Clim L(1) p 41

Keatinge WR, Donaldson GC (2006) Heat acclimatization andsunshine cause false indications of mortality due to ozone.Environ Res 100:387–393. doi:10.1016/j.envres.2005.08.012

Lambert WE, Samet JM, Dockery DW (1998) Community airpollution. In: Rom WN (ed) Environmental and occupationalmedicine. Lippincott-Raven, Philadelphia, pp 1501–1522

McGregor GR, Watkin HA, Cox M (2004) Relationships between theseasonality of temperature and ischaemic heart disease mortality:implications for climate based health forecasting. Clim Res25:253–263. doi:10.3354/cr025253

Medina-Ramón M, Schwartz J (2007) Temperature, temperatureextremes, and mortality: a study of acclimatisation and effectmodification in 50 US cities. Occup Environ Med 64:827–833

Nichols MC, Kalkstein LS, Cheng CS (1995) Possible human healthimpacts of a global warming. World Resour Rev 7:77–103

OMA (2005) Illness costs of air pollution. ICAP Summary Report.Ontario Medical Association, Toronto, p 11

Pirard P, Vandentorren S, Pascal M, Laaidi K, Le Tertre A, CassadouS, Ledrans M (2005) Summary of the mortality impactassessment of the 2003 heat wave in France. Euro Surveill10:153–156

Pope CA III, Kalkstein LS (1996) Synoptic weather modeling andestimates of the exposure–response relationship between dailymortality and particulate air pollution. Environ Health Perspect104:414–420. doi:10.2307/3432686

Qian Z, He Q, Lin HM, Kong L, Bentley CM, Liu W, Zhou D (2008)High temperatures enhanced acute mortality effects of ambient

particle pollution in the “oven” City of Wuhan, China. EnvironHealth Perspect 116:1172–1178

Rainham DGC, Smoyer-Tomic KE (2003) The role of air pollution inthe relationship between a heat stress index and human mortalityin Toronto. Environ Res 93:9–19. doi:10.1016/S0013-9351(03)00060-4

Rocklöv J, Forsberg B (2008) The effect of temperature on mortalityin Stockholm 1998–2003: a study of lag structures and heatwaveeffects. Scand J Public Health 36:516–523. doi:10.1177/1403494807088458

Ross T, Lott N (2003) A Climatology of 1980–2003 Extreme Weatherand Climate Events. National Climate Data Center TechnicalReport No. 2003-01, p 14

Schwartz J (2000) Assessing confounding, effect modification, andthresholds in the association between ambient particles and dailydeaths. Environ Health Perspect 108:563–568. doi:10.2307/3454620

Schwartz J, Spix C, Wichmann HE, Malin E (1991) Air pollution andacute respiratory illness in five German communities. EnvironRes 56:1–14. doi:10.1016/S0013-9351(05)80104-5

Sheridan SC, Kalkstein LS (2004) Progress in heat watch-warningsystem technology. Bull Am Meteorol Soc 85:1931–1941.doi:10.1175/BAMS-85-12-1931

Shin HH, Stieb DM, Jessiman B, Goldberg MS, Brion O, Brook J,Ramsay T, Burnett RT (2008) A temporal, multicity model toestimate the effects of short-term exposure to ambient airpollution on health. Environ Health Perspect 116:1147–1153

Stieb DM, Burnett RT, Beveridge RC, Brook JR (1996) Associationbetween ozone and asthma emergency department visits in SaintJohn, New Brunswick, Canada. Environ Health Perspect104:1354–1360. doi:10.2307/3432974

Toronto Public Health (2001) Toronto air quality index health linksanalysis. Technical Report. Toronto Public Health, Toronto, p 40

Toronto Public Health (2004) Air pollution burden of illness inToronto: 2004 summary. Technical Report. Toronto PublicHealth, Toronto, p 19

Vedal S, Brauer M, White R, Petkau J (2003) Air pollution and dailymortality in a city with low levels of pollution. Environ HealthPerspect 111:45–51

Wong CM, Vichit-Vadakan N, Kan H, Qian Z, the PAPA ProjectTeams (2008) Public health and air pollution in Asia (PAPA): amulticity study of short-term effects of air pollution on mortality.Environ Health Perspect 116:1195–1202

World Health Organization (2004) Health aspects of air pollution:results from the WHO project “systematic review of health aspectsof air pollution in Europe.” World Health Organization, ReportE83080. WHO Regional Office for Europe, Copenhagen, p 24

222 Air Qual Atmos Health (2008) 1:209–222